00:36 <urjaman> (also, the field is a float so to me it would seem odd to +1 it, but anyways...)

01:20 <alyssa> urjaman: Oops.

01:21 <alyssa> urjaman: MALI_NEGATIVE might be silly :)

07:25 <tomeu> narmstrong: btw, regarding https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2337 , you can just click the merge after CI button now that you have pushed the R-b tag

08:45 <narmstrong> tomeu: wasn’t aware of that, i’ll rebase and do that

09:24 <tomeu> wow, t720 is in a really bad state

09:24 <tomeu> narmstrong: is t820 also around the 3k passes ballpark?

09:28 <narmstrong> tomeu: yeah, they seem to be in a similar state

09:31 <narmstrong> tomeu: hmm maybe worse indeed

10:01 <tomeu> narmstrong: t720 worse than t820, right?

10:25 <warpme_> tomeu: playing with t720 + Xorg i'm constantly getting "AIGLX error: sun4i-drm does not export required DRI extension". Looking on xorg sources: https://cgit.freedesktop.org/xorg/xserver/tree/glx/glxdricommon.c?h=xorg-server-1.20.5#n351 I'm not so familiar with GLX to understand what is missing - but I'm getting this error only on t720 and not on t820/t860. It looks like for t720 soothing is missing....

11:03 <warpme_> shooting->something 😉

11:04 <narmstrong> tomeu: yep much worse, but t820 is already is an awful state !

11:12 <tomeu> warpme_: ok, thanks, let's see what I find

11:15 <warpme_> tomeu: thx. I'm on all most actual software stack (5.4-rc5; xorg1.20.5; libdrm2.4.100; and current git mesa). Hint me in any way If I can help...

11:49 <robmur01> T720 has the additional 'benefit' of all the largely-untested SFBD weirdness

11:51 <robmur01> although bizarrely, crusty old T620 actually seems less broken than T720 right now

12:37 <tomeu> alyssa: looks like the t720 doesn't like MALI_TILER_DISABLED

12:37 <tomeu> the blob uses the 0xfff hierarchy mask

12:37 <tomeu> with a size of 0x200

12:38 <tomeu> hmm, sounds familiar

12:41 <alyssa> Interesting

12:41 <alyssa> It's possible TILER_DISABLED is a newer feature

12:48 <tomeu> alyssa: though this is probably working on the T620

12:48 <alyssa> tomeu: Hm, yeah

12:48 <tomeu> so wonder if the T620 doesn't have some extra HW that the T720 doesn't

12:48 <tomeu> so userspace has to generate the polygon list in some cases

13:04 <alyssa> [ 953.957256] panfrost ff9a0000.gpu: AS_ACTIVE bit stuck

13:04 <alyssa> robher: ^ did I do a good??

13:25 <robher> alyssa: what triggered that?

13:26 <robher> I think only a page fault with runtime on inactive could cause that.

13:29 <alyssa> Genuinely unsure, just saw it in my demsg

13:29 <alyssa> dmesg

13:44 <tomeu> nice, got clears to work again

13:47 <tomeu> alyssa: 2 changes seem to be needed: enable checksum on scanout buffers and generate a polygon list when there's no draws

13:49 <robher> alyssa: is that on 5.3 or 5.4?

14:56 <tomeu> alyssa: I'm hitting this assert on t720 with kmscube on libmali: https://gitlab.freedesktop.org/mesa/mesa/blob/master/src/panfrost/encoder/pan_tiler.c#L261

14:58 <tomeu> alyssa: the hierarchy mask is 0x186

15:12 <tomeu> and when testing with a glmark2 scene, I fail at assert(mask & (0x80 | 0x100));

15:12 <tomeu> looks like we'll get to learn a bit more about the tiler :)

15:30 <alyssa> Wee

15:31 <alyssa> tomeu: Oh, very interesting.

15:32 <alyssa> The T860 blob doesn't do that, as I know

15:35 <alyssa> tomeu: Also to be clear -- checksumming is *optional*

15:35 <alyssa> So if enabling checksum fixes the bug something else is wrong.

15:35 <alyssa> Or the bits we thought enable the checksum do something else

15:41 <tomeu> yeah, no idea yet

20:08 <robmur01> well, that's one race bug down...

20:57 <urjaman> hm?

23:11 <alyssa> Weeeeeeeee

23:13 <HdkR> weeeeee

23:14 <anarsoul> alyssa: something works?

23:18 <alyssa> anarsoul: The race condition above